A personalized plan generation method based on full-chain air-rail intermodal travel

By constructing a classification model of random forest algorithm combined with decision trees, combining the in-city and intercity traffic information of air-rail joint journeys, a personalized travel plan is generated, which solves the problem that the existing air-rail joint journey plans cannot meet the needs of passengers and improves travel Selectability and service level.

CN113780396BActive Publication Date: 2024-11-29SOUTHEAST UNIV
View PDF 2 Cites 0 Cited by

Patent Information

Application Number
CN202111012909.5
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2021-08-31
Publication Date
2024-11-29
Estimated Expiration
2041-08-31

AI Technical Summary

Technical Problem

The existing air-rail joint travel plans lack personalization and cannot meet the real travel needs of passengers, resulting in the underutilization of market development potential.

Method used

Using a personalized scheme generation method based on full-chain air-rail connection, a random forest algorithm combined with a decision tree is constructed to determine the biggest influencing factors of passenger travel choices, and combine it with intra-city and inter-city. Transportation information, generate and sort travel plans, and provide personalized travel recommendations.

Benefits of technology

It improves the selectivity and service level of joint travel, meets the personalized needs of passengers, and enhances the attractiveness of air-rail joint travel.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN113780396B_ABST
    Figure CN113780396B_ABST
Patent Text Reader

Abstract

The present invention discloses a method for generating a personalized plan based on full-chain air-rail intermodal travel, which includes six steps: passenger travel data collection, determining the most influential factors of passenger travel choices, obtaining air-rail intermodal passenger travel needs, collecting intermodal travel information, generating and sorting travel plans, and formulating personalized travel plans. The present invention collects passengers' historical travel data and personal preference attributes, uses a random forest algorithm based on a decision tree to predict the most influential factors of different passengers' travel choices, considers passengers' full-chain air-rail intermodal travel needs, and customizes personalized intermodal travel plans for passengers.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention belongs to the field of multi-modal transportation and data mining, and specifically relates to a method for generating a personalized plan based on full-chain air-rail intermodal travel. Background Art

[0002] With the reform of the transportation supply side and the development of a modern integrated transportation system, urban transportation supply has shifted from a single road transportation network to a multi-modal integrated transportation network of "aviation network, road network, and rail network". Providing safe, fast, and punctual travel solutions has become an important task in the context of a "transportation power". However, there is currently a lack of effective cooperative management and control of aviation, high-speed rail, and highways. How to give full play to the advantages of multi-modal travel such as aviation, railways, and highways, avoid market chaos caused by disorderly competition, and build a full-chain multi-modal transportation system is an important research direction for the development of transportation in the new era.

[0003] In the generation of air-rail intermodal solutions, most studies discuss intercity transportation and intra-city transportation separately, lacking an overall consideration of the entire transportation chain. In addition, the development of air-rail intermodal in my country also faces problems such as an imperfect integrated ticketing information system and information barriers. In order to meet the real needs of intermodal passengers and take into account the heterogeneity of passengers, a personalized solution generation for the entire intermodal travel chain is created under the background of intercity transportation and intra-city transportation driving travel needs.

[0004] The current air-rail intermodal market has great development potential and meets the transportation development needs of the country and the people. However, the research on the personalized travel needs of passengers is not perfect enough, the travel options available to passengers cannot meet their real travel needs, and the attractiveness of air-rail intermodal travel to passengers is not ideal. Summary of the invention

[0005] Purpose of the invention: In view of the above problems, the present invention proposes a personalized plan generation method based on full-chain air-rail intermodal travel, which can fully consider the travel needs of different passengers, provide personalized intermodal travel plans, expand the air-rail intermodal selection space, and improve the intermodal travel service level.

[0006] Technical solution: To achieve the purpose of the present invention, the technical solution adopted by the present invention is: a method for generating a personalized plan based on a full-chain air-rail intermodal travel, comprising the following steps:

[0007] Step 1: Passenger travel data collection, including collecting historical travel data and perception data of N passengers within a year on air-rail connections within and between cities;

[0008] Historical travel data of the i-th passenger Including passenger income level A i 、Transportation before the journey B i、Passenger departure city C i 、Destination city of the passenger i , Transfer city E i 、Travel purpose of passengers i , the average distance travelled by passengers in one year G i , the average total cost of a passenger's trip in a year H i ;

[0009] The perception data of the i-th passenger Include departure time preferenceI i , Ticket purchase time preference J i 、The biggest influencing factor Y when passengers choose travel mode i , i=1,2,3…,N;

[0010] Step 2: Determine the factors that have the greatest impact on passengers' travel choices; divide the data of N passengers obtained in step 1 into a training set X train and the test set X test ;

[0011] Construct a classification model of the factors that have the greatest impact on passengers' travel choices, with ten attribute parameters A i ~J i As the independent variable of the model, the most influential factor Y i As the dependent variable of the model, the random forest algorithm combined with the decision tree algorithm was used to train the final classification model;

[0012] Step 3: Obtain the travel demand of air-rail connecting passengers, that is, the ten attribute parameters A of the passenger i ~J i ;

[0013] Step 4: Collect information about connecting trips; the trip consists of two parts: intra-city transportation and inter-city transportation. Intra-city transportation is provided by buses, while inter-city transportation is provided by aviation and high-speed rail. Collect flight prices P A , High-speed rail price P R 、City bus price P B , preferential subsidies for air and rail passengers S, flight duration T A , High-speed rail duration T R , transfer time T T 、Traffic time before and after the trip T C , and flight and high-speed rail schedules;

[0014] Step 5: Generate and sort travel plans: Based on the travel needs of passengers obtained in step 3, use the classification model trained in step 2 to obtain the factors that have the greatest impact on passengers' travel choices, query all feasible travel plans from departure date and departure city to destination city and sort them;

[0015] Step 6: Develop a personalized travel plan; this includes finding the most reasonable travel plan based on the factors that have the greatest impact on passengers' travel choices and sorting all travel plans in step 5, and completing personalized travel plan recommendations for air-rail connecting passengers.

[0016] Furthermore, the final classification model is obtained by training in step 2, and the method includes:

[0017] For the training set X train The random forest algorithm is used to randomly select T training subsets U with replacement T ;

[0018] In each subset, the data is classified according to the decision tree algorithm, the order of splitting nodes is determined according to the information gain ratio, and a decision tree is established;

[0019] For the test set X test All the data of , each set of data is substituted into the decision tree of T subsets separately, and T classification results of the factors that have the greatest impact on passengers' travel choices are obtained;

[0020] The random forest algorithm is used to vote on the classification results and count the voting results, and the dependent variable Y with the highest number of occurrences is used as the classification result of the data;

[0021] Test the classification model to determine the dependent variable Y predicted by the classification model i Whether it is consistent with the factors that have the greatest impact on the passenger's actual travel choice;

[0022] When the accuracy rate is higher than a certain threshold, it is determined that the classification effect meets the expected value and the model training ends. Otherwise, it returns to randomly select T subsets for classification training again, and repeats until the accuracy rate meets the requirements.

[0023] Furthermore, the order of splitting nodes is determined according to the information gain ratio, and the construction of the decision tree specifically includes:

[0024] 1) Calculate the training subset U T Information entropy Ent(U T ):

[0025]

[0026] In the formula, p k For the subset U T The proportion of samples of the kth category in U, k = 1, 2, 3, ..., n, n is T The number of sample types;

[0027] 2) Calculate subset U T The information gain of each attribute in Gain(U T ,a):

[0028] Suppose attribute A has V possible values ​​{A 1 ,A 2 ,…,A V}, where the value of attribute A is A V The sample size is U T V , the conditional probability distribution of attribute A is |U T V | / |U T |, then the information gain of attribute A Gain(U T ,A) is expressed as:

[0029] Gain(U T ,A)=Ent(U T )-Ent(U T , A)

[0030] Among them, Ent(U T , A) is the information entropy of attribute A;

[0031] 3) Traverse all attributes and obtain different information gains for different attributes;

[0032] 4) Calculate the information gain ratio g of all attributes R (U T , a);

[0033] For attribute A, the information gain ratio is g R (U T ,A), which is defined as information gain Gain(U T ,A) and training subset U T About the information entropy Ent(U T , A), that is

[0034] g R (U T , A)=Gain(U T , A) / Ent(U T , A)

[0035] 5) Traverse the information gain ratio g of all attributes R (U T ,a), find the maximum information gain ratio, and use the attribute corresponding to the maximum information gain ratio as the branch node;

[0036] 6) Loop step 2) to step 5) to find the attribute with the next maximum information gain ratio as the next branch node until all attributes are traversed.

[0037] Furthermore, the travel plan sorting method in step 5 is specifically as follows:

[0038] Calculate the total air-rail connection cost for each feasible travel plan: C = P A +P R +P B -S, total duration of passenger air-rail intercity travel T = T A +T R +T T +T C ;

[0039] Sort the feasible travel plans according to four methods: total cost from low to high, total time from low to high, departure time from early to late, and transfer time from low to high, to form a travel plan divided into four methods: total cost, total time, departure time, and transfer time.

[0040] Beneficial effects: Compared with the prior art, the technical solution of the present invention has the following beneficial technical effects: The personalized plan generation method based on the full-chain air-rail intermodal travel proposed by the present invention takes into account the personalized intermodal travel needs and actual intermodal travel information of passengers, and combines the dual drive of intra-city transportation and inter-city transportation to create a personalized plan recommendation system for the full-chain air-rail intermodal travel. This method increases the selectivity of travel plans for intermodal passengers on the one hand, and improves the service level of intermodal travel on the other hand, meeting the personalized travel needs of passengers. BRIEF DESCRIPTION OF THE DRAWINGS

[0041] Figure 1 It is a flow chart of the steps of the implementation of the present invention. DETAILED DESCRIPTION

[0042] The present invention will be further explained below in conjunction with the accompanying drawings and specific embodiments. It should be understood that the following specific embodiments are only used to illustrate the present invention and are not used to limit the scope of the present invention.

[0043] The present invention provides a personalized plan generation method based on full-chain air-rail intermodal travel, which provides a new idea for solving the personalized needs of air-rail intermodal passengers and generating full-chain travel plans. A database is established for the historical travel data of passengers, and a random forest algorithm based on a decision tree is trained to predict the most influential factors of passengers' travel choices. Finally, the personalized intermodal travel plan of passengers is customized from four perspectives: total cost, total duration, departure time, and transfer time, based on the intermodal travel information and personal preferences.

[0044] like Figure 1 As shown, a method for generating a personalized plan based on a full-chain air-rail intermodal travel disclosed in an embodiment of the present invention comprises the following steps:

[0045] Step 1: Passenger travel data collection; including collecting historical travel data and perception data of N passengers within a year regarding air-rail connections within the city and between cities, to determine the factors that have the greatest impact on passengers' travel mode selection.

[0046] Passenger travel dataX i Including passenger historical travel data and passengers’ perception data

[0047] Historical travel data of the i-th passenger Including passenger income level A i 、Transportation before the journey B i 、Passenger departure city C i 、Destination city of the passenger i , Transfer city E i 、Travel purpose of passengers i , the average distance travelled by passengers in one year G i , the average total cost of a passenger's trip in a year H i ;

[0048] The perception data of the i-th passenger Include departure time preferenceI i , Ticket purchase time preference J i 、The biggest influencing factor Y when passengers choose travel mode i .

[0049] In this embodiment, an SP survey is conducted on passengers using air-rail connections at Nanjing Lukou Airport, Nanjing South High-speed Railway Station, and Nanjing High-speed Railway Station, and the parameters are set as follows:

[0050] The income level of the i-th passenger is A i The value range is 1-5. i =1 means the average monthly income of the passenger is less than 5,000 yuan. i =2 means the average monthly income is 5001-10000 yuan, A i =3 means the average monthly income is 10001-15000 yuan, A i =4 means the average monthly income is 15001-20000 yuan, A i =5 means the average monthly income is more than 20,000 yuan;

[0051] The transportation method B of the i-th passenger i The value range is 1-4, B i =1 means public transportation, B i =2 means taking the subway, B i =3 means online car-hailing trip, B i =4 means traveling by private car;

[0052] The departure city of the i-th passenger is C i The value range is 1-5, C i =1 indicates first-tier cities, Ci =2 indicates a second-tier city, C i =3 indicates a third-tier city, C i =4 indicates a fourth-tier city, C i =5 indicates a fifth-tier city;

[0053] The destination city D of the i-th passenger i The value range is 1-5, D i =1 indicates first-tier cities, D i =2 indicates a second-tier city, D i =3 indicates a third-tier city, D i =4 indicates fourth-tier cities, D i =5 indicates a fifth-tier city;

[0054] The transfer city E of the i-th passenger i The value range is 1-6, E i =1 indicates first-tier cities, E i =2 indicates a second-tier city, E i =3 indicates a third-tier city, E i =4 indicates a fourth-tier city, E i =5 represents fifth-tier cities, E i =6 means no transfer;

[0055] The travel purpose of the i-th passenger F i The value range is 1-3, F i =1 means business trip, F i =2 means going to school / going home, F i =3 means travel;

[0056] The average distance G of a trip of the i-th passenger in a year i The value range is 1-4, G i =1 means less than 500 km, G i =2 means 500-1000 km, G i =3 means 1000-1500 km G i =4 means greater than 1500 km;

[0057] The average total cost of a trip of the i-th passenger in a year is H i The value range is 1-4, H i =1 means less than 500 yuan, H i =2 means 500-1000 yuan, H i =3 means 1000-2000 yuan, H i =4 means more than 2000 yuan;

[0058] The departure time preference of the i-th passenger I iThe value range is 1-4, I i =1 means 00:00-06:00, I i =2 means 06:00-12:00, I i =3 means 12:00-18:00, I i =4 means 18:00-24:00;

[0059] The i-th passenger's ticket purchase time preference J i The value range is 1-4, J i =1 means more than 15 days in advance, J i =2 means 7-15 days in advance, J i =3 means 3-7 days in advance, J i =4 means 0-3 days before departure.

[0060] Step 2: Determine the factors that have the greatest impact on passengers' travel choices; divide the data of N passengers obtained in step 1 into a training set X train and the test set X test ; Among them, the training set X train The number of is the value obtained by rounding N1=0.8*N. The test set X test The number of is N2=N-N1.

[0061] Construct a classification model of the factors that have the greatest impact on passengers' travel choices, with ten attribute parameters A i ~J i As the independent variable of the model, the most influential factor Y i As the dependent variable of the model, the random forest algorithm combined with the decision tree algorithm was used to train the final classification model.

[0062] The final classification model is trained by the following method:

[0063] For the training set X train The random forest algorithm is used, that is, T (T can be an odd number) training subsets U are randomly selected with replacement T (Random with replacement means that every time a sample is selected, it is equally likely to be selected again and added to the training set again);

[0064] In each subset, the data is classified according to the decision tree algorithm, the order of splitting nodes is determined according to the information gain ratio, and a decision tree is established;

[0065] For the test set X test All the data of , each set of data is substituted into the decision tree of T subsets separately, and T classification results of the factors that have the greatest impact on passengers' travel choices are obtained;

[0066] The random forest algorithm is used to vote on the classification results and count the voting results, and the dependent variable Y with the highest number of occurrences is used as the classification result of the data;

[0067] Test the classification model to determine the dependent variable Y predicted by the classification model i Whether it is consistent with the factors that have the greatest impact on the passenger's actual travel choice;

[0068] Assume that the number of consistent samples is m, the total number of samples is N2, and the accuracy rate is A=m / N2. When the accuracy rate is higher than 95%, it is determined that the classification effect meets the expected value and the model training is terminated. Otherwise, return to randomly select T subsets for classification training and repeat until the accuracy rate meets the requirements.

[0069] Among them, the order of splitting nodes is determined according to the information gain ratio, and the decision tree is constructed, which specifically includes:

[0070] 1) Calculate the training subset U T Information entropy Ent(U T ):

[0071]

[0072] In the formula, p k For the subset U T The proportion of samples of the kth category in U, k = 1, 2, 3, ..., n, n is T The number of sample types;

[0073] 2) Calculate subset U T The information gain of each attribute in Gain(U T ,a):

[0074] Suppose attribute A has V possible values ​​{A 1 , A 2 ,…,A V}, where the value of attribute A is A V The sample size is U T V , the conditional probability distribution of attribute A is |U T V | / |U T |, then the information gain of attribute A Gain(U T ,A) is expressed as:

[0075] Gain(U T ,A)=Ent(U T )-Ent(U T ,A)

[0076] Among them, Ent(U T ,A) is the information entropy of attribute A;

[0077] 3) Traverse all attributes and obtain different information gains for different attributes;

[0078] 4) Calculate the information gain ratio g of all attributes R (U T , a);

[0079] For attribute A, the information gain ratio is g R (U T ,A), which is defined as information gain Gain(U T ,A) and training subset U T About the information entropy Ent(U T , A), that is

[0080] g R (U T , A)=Gain(U T , A) / Ent(U T , A)

[0081] 5) Traverse the information gain ratio g of all attributes R (U T ,a), find the maximum information gain ratio, and use the attribute corresponding to the maximum information gain ratio as the branch node;

[0082] 6) Loop step 2) to step 5) to find the attribute with the next maximum information gain ratio as the next branch node until all attributes are traversed.

[0083] Step 3: For any passenger who needs to customize a personalized interline travel plan, obtain the travel needs of the air-rail interline passenger, that is, the ten attribute parameters A of the passenger i ~J i .

[0084] Step 4: Collect inter-city travel information; travel consists of two parts: intra-city travel and inter-city travel, and all travel is from the city center of one city to the city center of another city, which is a full-chain travel; intra-city travel is provided by buses, and inter-city travel is provided by aviation and high-speed rail;

[0085] Aviation data collection. Collect the departure city O requested by passengers on Ctrip, Qunar and other websites i , Transit City P i 、Destination City D i , departure time TO i As well as personal preference attributes, collect the flight prices P of all feasible routes from the departure city to the destination city A , preferential subsidies for air and rail passengers S, flight duration T A , High-speed rail duration T R, transfer time T T , and flight schedules.

[0086] Collection of high-speed rail data. Collect the high-speed rail prices P of all feasible routes from the departure city to the destination city of passenger i in the 12306 APP R 、City bus price P B , High-speed rail duration T R , transfer time T T , and the high-speed rail schedule.

[0087] In-city transportation data collection. In bus operator products such as Bus Butler that have established cooperation with airport high-speed rail stations, the price of in-city buses P B 、Traffic time before and after the trip T C , and bus schedules.

[0088] Step 5: Generate and sort travel plans. Based on the travel needs of passengers obtained in step 3, the classification model trained in step 2 is used to obtain the factors that have the greatest impact on passengers' travel choices. Then, travel apps such as Ctrip and 12306 are queried to query all feasible travel plans from departure date to departure city to destination city.

[0089] If passenger i chooses to take the high-speed rail first and then transfer to the plane, according to the feasible routes screened out, calculate the minimum total travel cost C = min (P R +P A +P B -S), the shortest total travel time of passenger i's air-rail connection is T = min (T T +T R +T A ++T C ), the earliest departure time is TO i = Schedule (city center - high-speed railway station), shortest transfer time T T =min(Transfer Time). Sort the feasible travel plans according to the total cost from low to high, the total time from low to high, the departure time from early to late, and the transfer time from low to high, and form a travel plan divided by the total cost, total time, departure time, and transfer time.

[0090] Step 6: Develop a personalized travel plan; this includes finding the most reasonable travel plan based on the factors that have the greatest impact on passengers' travel choices and sorting all travel plans in step 5, and completing personalized travel plan recommendations for air-rail connecting passengers.

[0091] The above description is only a preferred embodiment of the present invention and does not constitute any other form of limitation to the present invention. Any modification or equivalent change made based on the technical essence of the present invention still falls within the scope of protection required by the present invention.

Claims

1. A method for generating a personalized plan based on a full-chain air-rail intermodal travel, characterized in that: The method comprises the following steps: Step 1: Passenger travel data collection, including collecting historical travel data and perception data of N passengers within a year on air-rail connections within and between cities; Historical travel data of the i-th passenger Including passenger income level A i 、Transportation before the journey B i 、Passenger departure city C i 、Destination city of the passenger i , Transfer city E i 、Travel purpose of passengers i , the average distance travelled by passengers in one year G i , the average total cost of a passenger's trip in a year H i ; The perception data of the i-th passenger Include departure time preferenceI i , Ticket purchase time preference J i 、The biggest influencing factor Y when passengers choose travel mode i , i=1,2,3…,N; Step 2: Determine the factors that have the greatest impact on passengers' travel choices; divide the data of N passengers obtained in step 1 into a training set X train and the test set X test ; Construct a classification model of the factors that have the greatest impact on passengers' travel choices, with ten attribute parameters A i ~J i As the independent variable of the model, the most influential factor Y i As the dependent variable of the model, the random forest algorithm combined with the decision tree algorithm is used to train the final classification model. The process is as follows: For the training set X train The random forest algorithm is used to randomly select T training subsets U with replacement T ; In each subset, the data is classified according to the decision tree algorithm, the order of splitting nodes is determined according to the information gain ratio, and a decision tree is established, which includes: 1) Calculate the training subset U T Information entropy Ent(U T ): In the formula, p k For the subset U T The proportion of samples of the kth category in U, k = 1, 2, 3, ..., n, n is T The number of sample types; 2) Calculate subset U T The information gain of each attribute in Gain(U T ,a): Suppose attribute A has V possible values ​​{A 1 ,A 2 ,…,A V }, where the value of attribute A is A V The sample size is U T V , the conditional probability distribution of attribute A is |U T V | / |U T |, then the information gain of attribute A Gain(U T ,A) is expressed as: Gain(U T ,A)=Ent(U T )-Ent(U T ,A) Among them, Ent(U T ,A) is the information entropy of attribute A; 3) Traverse all attributes and obtain different information gains for different attributes; 4) Calculate the information gain ratio g of all attributes R (U T ,a); For attribute A, the information gain ratio is g R (U T ,A), which is defined as information gain Gain(U T ,A) and training subset U T About the information entropy Ent(U T ,A), that is g R (U T ,A)=Gain(U T ,A) / Ent(U T ,A) 5) Traverse the information gain ratio g of all attributes R (U T ,a), find the maximum information gain ratio, and use the attribute corresponding to the maximum information gain ratio as the branch node; 6) Loop through steps 2) to 5) to find the attribute with the next maximum information gain ratio as the next branch node until all attributes are traversed; For the test set X test All the data of , each set of data is substituted into the decision tree of T subsets separately, and T classification results of the factors that have the greatest impact on passengers' travel choices are obtained; The random forest algorithm is used to vote on the classification results and count the voting results, and the dependent variable Y with the highest number of occurrences is used as the classification result of the data; Test the classification model to determine the dependent variable Y predicted by the classification model i Whether it is consistent with the factors that have the greatest impact on the passenger's actual travel choice; When the accuracy rate is higher than a certain threshold, it is determined that the classification effect meets the expected value and the model training ends. Otherwise, it returns to randomly select T subsets for classification training again, and repeats until the accuracy rate meets the requirements. Step 3: Obtain the travel demand of air-rail connecting passengers, that is, the ten attribute parameters A of the passenger i ~J i ; Step 4: Collect information about connecting trips; the trip consists of two parts: intra-city transportation and inter-city transportation. Intra-city transportation is provided by buses, while inter-city transportation is provided by aviation and high-speed rail. Collect flight prices P A , High-speed rail price P R 、City bus price P B , preferential subsidies for air and rail passengers S, flight duration T A , High-speed rail duration T R , transfer time T T 、Traffic time before and after the trip T C , and flight and high-speed rail schedules; Step 5: Generate and sort travel plans. Based on the travel needs of passengers obtained in step 3, the classification model trained in step 2 is used to obtain the factors that have the greatest impact on passengers' travel choices. All feasible travel plans from departure date to departure city to destination city are queried and sorted, as follows: Calculate the total air-rail connection cost for each feasible travel plan: C = P A +P R +P B -S, total duration of passenger air-rail intercity travel T = T A +T R +T T +T C ; Sort the feasible travel plans according to the four methods of total cost from low to high, total time from low to high, departure time from early to late, and transfer time from low to high, and form travel plans divided by total cost, total time, departure time, and transfer time; Step 6: Develop a personalized travel plan; this includes finding the most reasonable travel plan based on the factors that have the greatest impact on passengers' travel choices and sorting all travel plans in step 5, and completing personalized travel plan recommendations for air-rail connecting passengers.

Citation Information

Patent Citations

  • Passenger travel multivariate data analysis-based sky train joint transit place determination method

    CN110879866A

  • Random forest sudden fault early warning method based on sliding window

    CN112966434A